本视频为AI带路党Pro视频如何在Dify沙盒中安装运行pandas、numpy-深入Dify沙盒原理-Dify深入学习系列2 配套代码，介绍如何在Dify工作流中的代码执行节点（Code）运行pandas，并以此为切入点介绍Dify sandbox的原理和如何安装其他包
需要结合dify源码使用
在dify的docker目录中的docker-compose.yaml文件中，修改sandbox的配置中的volumes配置，确保配置为
volumes:
- ./volumes/sandbox/dependencies:/dependencies
- ./volumes/sandbox/conf:/conf然后将项目中的conf和dependencies的文件拷贝进入dify->docker->volumes->sandbox目录中
docker compose up启动后，进入docker-sandbox-1容器中，进入/dependencies/code目录中执行
bash test.sh 将获得的缺少的系统调用拷贝出来，和默认的系统调用合并
可以根据 cpu 架构不同在系统中/dependencies/code目录下执行 get_default_syscalls_amd64或get_default_syscalls_arm64获取默认系统调用
例如arm64
./get_default_syscalls_arm64
输出默认系统调用示例: >293,64,57,56,63,62,61,98,222,214,226,215,139,135,132,216,146,144,174,172,173,178,93,94,131,134,29,124,100,99,293,20,113,169,101,21,115,72,85,86,87,278,220,34,198,203,200,201,202,206,207,212,209,204,205,208,73,160,269,79,80,44,22
或者你也可以使用我测试得出的默认调用
amd64 默认系统调用
318,334,307,262,16,8,217,1,3,257,0,202,9,12,10,11,15,25,105,106,102,39,110,186,60,231,234,13,16,24,273,274,334,228,96,35,291,233,230,270,201,14,131,318,56,258,83,41,42,49,50,43,44,45,51,47,52,54,271,63,46,307,55,5,72,138,7,281
arm64 默认系统调用
293,64,57,56,63,62,61,98,222,214,226,215,139,135,132,216,146,144,174,172,173,178,93,94,131,134,29,124,100,99,293,20,113,169,101,21,115,72,85,86,87,278,220,34,198,203,200,201,202,206,207,212,209,204,205,208,73,160,269,79,80,44,22
类似conf/config.yaml中配置 修改
allowed_syscalls: [...合并后的系统调用数组]
最后重启,docker compose down 在docker compose up即可调用
curl -X POST http://localhost:8194/v1/sandbox/run -H "X-Api-Key: dify-sandbox" -H "Content-Type: application/json" -d '{"language":"python3","preload":"preload","enable_network":true,"code":"import pandas as pd"}'